JPH10224803A - Motion adaptive nonlinear quantizer and motion compensated inter-frame coder using the same - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide the motion adaptive nonlinear quantizer where oscillation between noise frames is suppressed and to provide the motion compensated inter-frame coder where image quality is improved by adding a quantization pre-processing to the motion compensation inter-frame coder having been already developed. SOLUTION: A difference signal S outputted from a difference device 9 is quantized by a nonlinear quantization section 5. The nonlinear quantization section 5 shows quantization characteristics of an inverse MAX nonlinear quantizer. A quantization level number Q outputted from the nonlinear quantization section 5 is given to a variable length coder 11 and an inverse quantizer 6. Since the inverse MAX nonlinear quantizer has a characteristic that a quantization step size is larger when the difference signal S is small and the quantization step size is gradually smaller when the difference signal S is larger, production of quantization noise at a fast motion part of the image is suppressed.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a motion-adaptive nonlinear quantizer and a motion-compensated interframe coding apparatus using the same.
The present invention relates to a motion-adaptive nonlinear quantizer capable of efficiently reducing the amount of information, and a motion-compensated interframe coding device using the same.

[0002]

2. Description of the Related Art A quantizer used in a conventional motion-compensated interframe coding apparatus is a linear quantizer having characteristics as shown in FIG. 11 or 0 based on the linear quantizer. A quantizer with a dead zone having a characteristic as shown in FIG. 12 for slightly increasing the step size in the vicinity has been used. The horizontal axis S in FIGS. 11 and 12 indicates the input value,
The vertical axis Q indicates the quantization level number.

[0003]

As is well known, human visual characteristics are insensitive to noise having no or small vibration between frames, and are sensitive to noise having large vibration. . In the motion-compensated interframe coding device,
The difference signal between the current image and the previous image is the input value S,
Since the linear quantizer has the same step size regardless of the size of the input value S, there is a problem that quantization noise is generated in a portion where the image moves fast and is superimposed on the image. Further, the quantizer with the dead zone has an advantage that the amount of information can be suppressed without deteriorating the image quality because the step size is large when the input value S is small. However, there is a problem that quantization noise is generated in a portion where an image moves fast, as in the case of the linear quantizer. Further, there is a problem that noise superimposed on the image vibrates between frames of the image and gives a great sense of discomfort visually.

[0004] It is an object of the present invention to provide a motion adaptive nonlinear quantizer which can eliminate the above-mentioned problems of the prior art and can suppress noise between frames of noise. Another object is to provide a motion-compensated inter-frame coding apparatus with improved image quality by adding a pre-quantization process to a motion-compensated inter-frame coding apparatus already developed. is there.

[0005]

SUMMARY OF THE INVENTION In order to achieve the above object, the present invention provides a motion-adaptive nonlinear quantum system for reducing the amount of information in a moving image in consideration of visual characteristics in a time direction. An inverse MAX type nonlinear quantizer applied to a difference signal between a current coded frame and a motion-compensated predicted frame, and a nonlinear characteristic controlling a quantization characteristic of the inverse MAX type nonlinear quantizer. The first feature resides in that a control unit is provided.

Further, in a motion compensation inter-frame coding apparatus on the encoding side having a motion compensation inter-frame DPCM loop, a prediction just before a quantization representative value is generated in the motion compensation inter-frame DPCM loop on the encoding side. Error is calculated by inverse M
A second feature is that a nonlinear quantizer for quantizing with the AX type nonlinear quantization characteristic is provided.

[0007] According to the present invention, the current coded frame,
When the value of the difference signal between the motion-compensated prediction frames is small, the quantization step size becomes large, and when the value of the difference signal is large, the quantization step size becomes small. It is possible to achieve the reduction and the suppression of the generation of the quantization noise in the fast moving part of the image.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below in detail with reference to the drawings. FIG. 1 is a block diagram showing a configuration of an embodiment of a motion compensation interframe coding apparatus to which a motion adaptive nonlinear quantizer according to the present invention is applied.

In FIG. 1, reference numeral 1 denotes a motion detecting unit which performs motion detection in units of, for example, blocks of 8 pixels × 8 lines. Reference numeral 2 denotes a frame memory for storing image data of a previous frame in order to perform motion detection. Reference numeral 3 denotes a motion correction unit that corrects motion in units of blocks. 4
Is a quantization characteristic control unit for controlling the quantization characteristic. Reference numeral 5 denotes a non-linear quantization unit, and reference numeral 6 denotes an inverse quantization unit. Reference numeral 7 denotes an orthogonal transform unit such as a DCT, and reference numeral 8 denotes an inverse orthogonal transform unit. Reference numeral 9 denotes a differentiator between the motion-compensated predicted frame signal and the frame, and 10 denotes an adder for obtaining a locally decoded signal. 11 is a variable length encoder, 12 is a buffer for temporarily storing the signal encoded by the variable length encoder 11, and 13 is a quantization step size (Δ) based on the amount of data stored in the buffer 12. Is a quantization step size (Δ) determination unit for determining the quantization step size. A well-known quantization step size (Δ) determination unit 13 can be used. FIG. 2 is a block diagram showing a specific example of the nonlinear quantization unit 5. This specific example is suitable when the inverse quantizer used in the decoder not shown and the inverse quantizer 6 in FIG. 1 are based on linear inverse quantization. is there.

As shown, the nonlinear quantization section 5 comprises a conversion table 21 for converting the output signal S of the orthogonal transformation section 7 into S ', and a rounding-off processing section 22. . The round-off processing unit 22 includes a division unit 22a, an addition unit 22b, and a truncation unit 22c. As a conversion equation used in the conversion table 21, for example, the following equation (1) can be used.

[0011]

(Equation 1) Here, Δ is the quantization step size (Δ) determination unit 1
3 shows the quantization step size output from No. 3. Also,
a is a quantization characteristic determined by the quantization characteristic control unit 4.

An example of how to determine the quantization characteristic a will be described below. The quantization characteristic control unit 4 stores orthogonal transform coefficient position information (horizontal m = 1, 2,..., 8, vertical n =
1, 2,..., 8), coding mode information (M = I, P, B
3 types of pictures) and macroblock activity information (A; F = flat, E = edge, D = fine detail). Then, the quantization characteristic control unit 4 determines the quantization characteristic a (m, n, M, A) from these information from the following equations (2) to (4-3). Note that these formulas are merely examples, and the present invention is not limited to these. (1) When M = I, for any m, n, A:

A (m, n, M, A) = 0 (2) (2) When M = P, a (m, n, P, F) = 3 (m ² + n ² ) ^0.27 . -1) a (m, n, P, E) = 2 (3-2) a (m, n, P, D) = 4 (m ² + n ² ) ^0.17 (3-3) (3) M When = B, a (m, n, B, F) = 5 (m ² + n ² ) ^0.27 (4-1) a (m, n, B, E) = 3 (4-2) a ( (m, n, B, D) = 7 (m ² + n ² ) ^0.27 (4-3) The output signal S ′ converted by the equation (1) is rounded off by the rounding-off processing section 22. And sent to the subsequent circuit.

According to the above embodiment, the quantization characteristic from the orthogonal transform output signal S to the quantization level number Q, that is, the quantization characteristic of the non-linear quantizer 5 is I, P and B picture, respectively. 3, 4, and 5. 3, 4, and 5, the horizontal axis represents the input value S, and the vertical axis represents the quantization level number Q. In these figures, the curve a
Indicates a characteristic when the quantization step size Δ is small, and a curve b indicates a characteristic when the quantization step size Δ is large.

In the case of an I picture, a (m, n, M, A)
Since = 0, a linear quantization characteristic as shown in FIG. 3 is obtained.
In the case of a P picture, when the input value S is small, the quantization step size Δ ′ is large, but when the input value S is large, the quantization step size Δ ′ tends to be small.
This tendency is amplified as the quantization step size Δ increases, as is clear from the curves a and b. Further, in the case of a B picture, the quantization characteristic a is P
Since it is larger than that of the picture, it can be seen that the nonlinear characteristic is further amplified.

As is apparent from FIGS. 4 and 5, the input value S
When the orthogonal transformation output signal is small, the quantization step size Δ ′ is large, but when the input value S increases, the quantization step size Δ ′ gradually decreases. This characteristic can be called an inverse MAX type nonlinear characteristic.

According to this embodiment, when the input value S is small, the amount of information can be suppressed because the quantization step size Δ 'is large, and when the input value S is large, the quantization step size Δ' is small. In addition, it is possible to suppress the occurrence of quantization noise in a fast-moving part of an image. As a result, noise between frames can be suppressed.

As described above, the non-linear quantization unit 5 having the above-described configuration is used when the inverse quantization unit of the decoder and the inverse quantization unit 6 in FIG. 1 are based on linear inverse quantization. Can effectively suppress the vibration between noise frames. However, when an inverse quantizer for MPEG-2 decoding is used as the inverse quantization unit, the result becomes as shown in FIG. Some failures occur.

That is, the inverse quantization characteristic of the inverse quantizer for MPEG-2 decoding can be expressed as follows. Now, assuming that the quantized reproduction value of the inverse quantizer for MPEG-2 decoding is F, the quantization level number input to the inverse quantizer for MPEG-2 decoding is Q, and the quantization step size is Δ, The quantized reproduction value F can be expressed by the following equation. F = (2Q + Sign (Q)) × Δ / 2 Here, Sign (Q) is 1 when Q> 0, 0 when Q = 0, Q
<0 when −1. In the first embodiment described above, 4
Since the rounding processing unit 22 performs the rounding processing, the quantization level number Q generated by the quantizer on the encoder side is used.
Will be inversely quantized by the inverse quantizer for MPEG-2 decoding as shown in FIG. That is,
0.5 ≦ Δ <1.5 of the quantization level number Q on the encoder side
In the inverse quantizer for MPEG-2 decoding, the quantized reproduced value F = 1.5, and 1.5 ≦ Δ <2.5 is the quantized reproduced value F = 2.5, 2.5 ≦ Δ < 3.5 is the quantized reproduction value F =
.., And the quantization level number Q on the encoder side
And the quantization reproduction value F is not well-matched. The second embodiment of the present invention solves this problem.

The second embodiment is characterized in that the configuration of FIG. 7 is used as the nonlinear quantizer of FIG. In the figure, 31 is a conversion table, and 32 is a truncation processing unit. The truncation processing unit 32 includes a division unit 32a
And a cut-off section 32b. As specific examples of the conversion table 31, the following equations (5-1) and (5-2) can be raised.

[0021]

(Equation 2) Note that as a, the above formulas (2), (3-1) to (3-3) and (4-1) to (4-3) can be used.

[0022] The using the equation (5-1) with a · Δ ^2/48, in the MPEG-2 quantizer, the quantization step size (delta) determination section 13, already a dead zone characteristic This is to prevent the nonlinear quantization unit 5 from exhibiting an excessive dead zone characteristic.

When the above equations (5-1) and (5-2) are used, M
FIG. 8 shows the relationship between the quantized reproduction value F of the inverse quantizer for PEG-2 decoding and the quantization level number Q input to the inverse quantizer for MPEG-2 decoding. As is clear from FIG. 8, 1.0 ≦ Δ <of the quantization level number Q on the encoder side.
In the inverse quantizer for MPEG-2 decoding, 2.0 is the quantized reproduction value F = 1.5, and 2.0 ≦ Δ <3.0 is the quantized reproduction value F = 2.5, 3.0. .Ltoreq..DELTA. <4.0 results in a quantized reproduction value F = 3.5,...
The problem of the embodiment can be solved.

The quantization characteristics of this embodiment are different from those of FIGS. 3, 4 and 5 in that the quantization step size Δ at the input value S = 0 is about twice as large as that of FIGS. The same is true of the point showing the type nonlinear characteristic. Therefore, it is natural that the second embodiment can also achieve the effects of the first embodiment.

[0025]

As is apparent from the above description, according to the present invention, the quantizer having the inverse MAX type nonlinear characteristic is applied to the difference signal between the current coded frame and the motion-compensated predicted frame. Therefore, it is possible to suppress the generation of quantization noise in a portion where the motion of the image is fast. As a result,
It becomes possible to suppress noise between frames of noise.

Further, a motion adaptive nonlinear quantizer suitable for use in the MPEG-2 quantizer can be provided.

Under the test data and coding parameters shown in FIG. 9, quantization was performed using the following three types of quantization methods. Method 1: Linear quantizer + linear inverse quantizer (conventional method) Method 2: MPEG-2 Linear quantizer with dead zone +
MPEG-2 inverse quantizer (conventional method) method 3 ... inverse MAX type non-linear quantizer + linear quantizer or MPEG-2 inverse quantizer (method of the present invention) Quantized using methods 1 to 3 above However, the result as shown in FIG. 10 was obtained. From FIG. 10, according to the method of the present invention, good results were obtained in five-step evaluations for both the flower garden and the mobile and calendar test data. In the subjective evaluation, the mosquito noise was greatly reduced in the flower garden, and the effect of eliminating the noise was obtained. In the mobile and calendar, the deterioration of the characters and the edge portion was significantly reduced.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of an embodiment of a motion compensation interframe coding apparatus to which a motion adaptive nonlinear quantizer according to the present invention is applied.

FIG. 2 is a block diagram showing a configuration of an embodiment of a motion adaptive nonlinear quantizer according to the present invention.

FIG. 3 is a diagram showing a tendency of a quantization characteristic in an I picture of the nonlinear quantizer of the embodiment.

FIG. 4 is a diagram showing a tendency of a quantization characteristic in a P picture of the nonlinear quantizer of the embodiment.

FIG. 5 is a diagram illustrating a tendency of a quantization characteristic in a B picture of the nonlinear quantizer of the embodiment.

FIG. 6 is a diagram illustrating a relationship between a quantization level number Q obtained by the first embodiment and a quantization reproduction value F of an inverse quantizer for MPEG-2 decoding.

FIG. 7 is a block diagram illustrating a configuration of a motion adaptive nonlinear quantizer according to a second embodiment of the present invention.

FIG. 8 is a diagram illustrating a relationship between a quantization level number Q obtained by the second embodiment and a quantized reproduction value F of an inverse quantizer for MPEG-2 decoding.

FIG. 9 is a diagram illustrating an example of test data and encoding parameters.

FIG. 10 is an explanatory diagram for comparing image quality obtained by a conventional method and a method of the present invention.

FIG. 11 is a diagram illustrating a characteristic example of a conventional linear quantizer.

FIG. 12 is a diagram illustrating a characteristic example of a conventional quantizer with a dead zone.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Motion detection part, 2 ... Frame memory, 3 ... Motion correction part, 4 ... Quantization characteristic control part, 5 ... Non-linear quantization part, 6 ...
Inverse quantization unit, 7: orthogonal transform unit, 8: inverse orthogonal transform unit, 9:
Differentiator, 10 adder, 11 variable length encoder, 12
Buffer, 13 ... quantization step size (Δ) determination unit,
21, 31... Conversion table, 22.
2 ... Truncation processing unit.

Claims (7)

[Claims] 1. A motion-adaptive nonlinear quantizer for reducing the amount of information in a moving image in consideration of visual characteristics in a time direction, wherein a current coded frame and a motion-compensated predicted frame are Motion-adaptive nonlinear quantization, comprising: an inverse MAX type nonlinear quantizer applied to the differential signal of the above; and a nonlinear characteristic control unit for controlling a quantization characteristic of the inverse MAX type nonlinear quantizer. vessel. 2. The motion adaptive nonlinear quantizer according to claim 1, wherein the inverse MAX nonlinear quantizer includes a quantization step size determined based on an encoded information amount of the difference signal; A motion adaptive nonlinear quantizer, comprising: a conversion table for generating an inverse MAX nonlinear quantization characteristic by a control signal obtained from a characteristic control unit. 3. The motion adaptive nonlinear quantizer according to claim 2, wherein the inverse MAX nonlinear quantizer rounds off a decimal part of a signal obtained from the conversion table. A motion-adaptive nonlinear quantizer, comprising a truncation unit or a truncation unit for truncation. 4. The motion-adaptive nonlinear quantizer according to claim 1, wherein the nonlinear characteristic control unit calculates the inverse MAX type from orthogonal transform coefficient position information, coding mode information, and macroblock activity information. A motion adaptive nonlinear quantizer for obtaining the control signal a for controlling a quantization characteristic of the nonlinear quantizer. 5. A motion-compensated inter-frame coding apparatus having a motion-compensated inter-frame DPCM loop, comprising: a motion vector detecting means for locally detecting a motion vector with respect to a moving image; A motion compensator for performing motion compensation based on the obtained motion vector; an inverse MAX nonlinear quantizer applied to a difference signal between the current coded frame and the motion compensated predicted frame; A motion-compensated interframe coding apparatus, comprising: a non-linear characteristic control unit for controlling a quantization characteristic of a non-linear quantizer. 6. An encoding-side motion-compensation inter-frame coding apparatus having a motion-compensation inter-frame DPCM loop, comprising:
A motion-compensated interframe coding apparatus, comprising: a non-linear quantizer that quantizes a prediction error immediately before a quantization representative value is generated by an inverse MAX type non-linear quantization characteristic. 7. The motion-compensated interframe coding apparatus according to claim 6, wherein a quantization step size obtained from the quantizer is determined by using a non-linear quantizer. A motion-compensated interframe coding apparatus, wherein one parameter for determining a quantization characteristic is used.